Making the Accounting System Answerable by an AI Assistant

The challenge

Answering an ordinary business question — what did we spend with this supplier, what is outstanding on this job — meant finding someone who knew where to look in the accounting system and asking them to run it. The data was not missing; the path to it ran through a person.

Where we came in

Bolting a chatbot onto an ERP is the wrong shape: it hard-codes one assistant to one system and ages badly. The Model Context Protocol is an emerging open standard for exactly this, so we implemented against the standard rather than against a specific assistant, which means the integration outlives whichever AI tool the business settles on.

What we did

• Implemented a Model Context Protocol interface over the accounting system rather than a bespoke chatbot integration
• Scoped access through directory app registrations so the interface inherits existing permissions
• Exposed the business entities that matter operationally rather than the entire data model
• Built against an open standard so the same integration serves whichever assistant the business adopts

Where it landed

• Business data in the accounting system can be queried through AI tooling rather than through a report request
• Built against an open protocol, so it is not tied to a single AI vendor
• Access runs through existing directory permissions rather than a separate credential